# Clean-room reproducibility report

Status: `PASS`

- Started: `2026-08-03T14:02:39.991050+00:00`
- Completed: `2026-08-03T14:04:59.971687+00:00`
- Workspace: `<cleanroom_workspace>`
- Checks: `16/16 PASS`

## Scope

The path-isolated rebuild starts from direct GEO matrices/tables for GSE233606,
the GSE272233 raw count matrix, the GSE277637 raw tar, and newly regenerated
sample/donor pseudobulks from the remote PRJNA772047, PRJNA1218493 and GSE156497
source objects. It then rebuilds all analysis layers used by NAR Figure 8.

## Output agreement

| Object | Comparison | Status | Detail |
| --- | --- | --- | --- |
| GSE233606 condition effects | numeric_table | PASS | 62,703 rows |
| GSE233606 unselected metrics | numeric_table | PASS | 1 rows |
| GSE272233 correction vector metrics | numeric_table | PASS | 3 rows |
| GSE272233 camera tests | numeric_table | PASS | 9,354 rows |
| GSE272233 reversal consensus | numeric_table | PASS | 1,559 rows |
| External pathway effects | numeric_table | PASS | 4,663 rows |
| External pathway consensus | numeric_table | PASS | 1,629 rows |
| External sample QC | numeric_table | PASS | 10 rows |
| GSE277637 heterogeneity metrics | numeric_table | PASS | 32 rows |
| Figure 8 source data | numeric_table | PASS | 20 rows |
| Dataset statistical units | numeric_table | PASS | 6 rows |
| Candidate DMD truth boundary | numeric_table | PASS | 21 rows |
| GSE272233 Reactome summary | selected_json_fields | PASS | design; results; claim_boundary |
| External pathway summary | selected_json_fields | PASS | datasets; results; claim_boundary |
| NAR measured DMD summary | selected_json_fields | PASS | candidate_count; measured_candidate_gene_dmd_perturbations; candidate_level_dmd_pathway_predictions; claim_boundary |
| Figure 8 raster pixels | rgba_pixel_sha256 | PASS | metadata-independent pixel comparison |

## Important limitations

- The local `GSE233606_RAW.tar` is truncated. This run uses the complete direct
  GEO supplementary MatrixMarket, cell and feature files and records their hashes.
- The remote raw-object-to-pseudobulk step ran on the project Bohrium host before
  the compact outputs entered the isolated local workspace.
- Python and R packages are system-provided. This is path/process isolation, not
  a container-level or independently administered reproduction.
- Computational agreement does not establish biological validity, candidate-level
  DMD response truth or therapeutic efficacy.